I think the core idea is that it actually gives you an inverse function, based on your regular (in the DSL) function definition.
There's as far as I can tell no inherent need to call the regular function first - given any function f, you get a callable inverse function ~f.